Crystal structure of carbapenem synthase (CarC).

The proposed biosynthetic pathway to the carbapenem antibiotics proceeds via epimerization/desaturation of a carbapenam in an unusual process catalyzed by an iron- and 2-oxoglutarate-dependent oxygenase, CarC. Crystal structures of CarC complexed with Fe(II) and 2-oxoglutarate reveal it to be hexameric (space group C2221), consistent with solution studies. CarC monomers contain a double-stranded beta-helix core that supports ligands binding a single Fe(II) to which 2-oxoglutarate complexes in a bi-dentate manner. A structure was obtained with l-N-acetylproline acting as a substrate analogue. Quantum mechanical/molecular mechanical modeling studies with stereoisomers of carbapenams and carbapenems were used to investigate substrate binding. The combined work will stimulate further mechanistic studies and aid in the engineering of carbapenem biosynthesis.

The proposed biosynthetic pathway to the carbapenem antibiotics proceeds via epimerization/desaturation of a carbapenam in an unusual process catalyzed by an iron-and 2-oxoglutarate-dependent oxygenase, CarC. Crystal structures of CarC complexed with Fe(II) and 2-oxoglutarate reveal it to be hexameric (space group C222 1 ), consistent with solution studies. CarC monomers contain a double-stranded ␤-helix core that supports ligands binding a single Fe(II) to which 2-oxoglutarate complexes in a bi-dentate manner. A structure was obtained with L-N-acetylproline acting as a substrate analogue. Quantum mechanical/molecular mechanical modeling studies with stereoisomers of carbapenams and carbapenems were used to investigate substrate binding. The combined work will stimulate further mechanistic studies and aid in the engineering of carbapenem biosynthesis.
Carbapenems possess a broad spectrum of antibacterial activity and are relatively stable to serine ␤-lactamases that are a major cause of resistance to penicillins and cephalosporins (1,2). Carbapenems, such as thienamycin, were first isolated from bacterial extracts, including those from Serratia marcescens, Erwinia carotovora, and Streptomyces cattleya (3,4). Early attempts to improve the fermentation titers to commercially useful levels were unsuccessful (5), and carbapenem use is, in part, limited by production costs. Synthetic methodology for carbapenem production has been developed (6) but is less efficient than the direct fermentation or "semi-synthetic" procedures used for the production of penicillins and cephalosporins. Studies on carbapenem biosynthesis are of interest because they may enable engineering of the pathway to produce either medicinally useful antibiotics directly or intermediates for their production.
Bycroft, Salmond, and coworkers (7) have discovered that the low fermentation titers of carbapenems are due to the operation of a "quorum sensing" machinery in their regulation, open-ing the way to increased fermentation yields. The sequences of the nine genes (in E. carotovora) responsible for the biosynthesis of (5R)-carbapenem, the simplest known natural carbapenem have been reported (8). Although this compound is not a medicinally useful carbapenem due to its lack of a C-6 side chain, it nonetheless shares an identical nucleus with carbapenems that are (4). Eight of the genes, carA-H, are organized as an operon controlled by CarR, an LuxR type transcriptional activator (9,10). Five of the genes, carA-E, are believed to be directly involved in the production of the (5R)-carbapenem nucleus, and it has been shown that it can be produced in Escherichia coli, albeit at lower levels, using only CarA-C (11,12).
It has been proposed that glutamate semi-aldehyde, formed by the action of CarD and CarE, condenses with acetyl-CoA in a CarB-catalyzed reaction to give a monocyclic pyrolidine intermediate (5S-carboxymethyl)-S-proline (trans-CMP) (Scheme 1) (2,13). This can then be cyclized in an ATP-driven process catalyzed by CarA, to give a (3S,5S)-carbapenam (14). This reaction is closely related to ␤-lactam formation during biosynthesis of the ␤-lactamase inhibitor clavulanic acid (15)(16)(17) but is very different from ␤-lactam formation during penicillin biosynthesis (18). In this proposed pathway and consistent with the reports of both Bycroft et al. (14) and Li et al. (12), the role of CarC is to catalyze both epimerization at C-5 and desaturation across the C-2/C-3 bond of the carbapenam.
CarC is related to clavaminic acid synthase (CAS) 1 (approximately 23% sequence identity) (8), which catalyzes three reactions, comprising hydroxylation, ring closure, and desaturation processes, during clavulanic acid biosynthesis (22). Although the desaturation process catalyzed by CarC follows the precedent set by the CAS desaturation reaction, its assigned epimerization reaction is unique. To understand the mechanism of the highly unusual CarC reactions, structural information is required. Here we report the in vitro enzyme-mediated production of an active carbapenem antibiotic and describe crystal structures of CarC that define its active site structure thus enabling mechanistic and engineering studies aimed at altering its selectivity.

EXPERIMENTAL PROCEDURES
CarC Cloning, Expression, and Purification-A PCR-amplified DNA product corresponding to the E. carotovora carC gene (8) was engineered as an NdeI-BamHI fragment into the pET24a expression vector (Novagen) and transformed into E. coli BL21(DE3) supercompetent cells. Cells were grown in shake flasks at 37°C using 2TY medium containing 50 g/ml kanamycin. At A 600 0.8, cells were induced with 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside and growth allowed to continue for 4 h, CarC expression was ϳ20% of the total soluble protein. Cultured cells were resuspended in 50 mM Tris-HCl, pH 8.0, 1 mM dithiothreitol, 2 mM EDTA, 100 mM NaCl, 0.16 mg/ml lysozyme, with 8 l phenylmethylsulfonyl fluoride per gram of cell pellet (50 mM phenylmethylsulfonyl fluoride stock made up in 100% isopropanol) and 0.05% polyethyleneimine. Suspended cells were sonicated (HeatSystems), and the lysate was centrifuged (35,000 ϫ g) for 30 min. The resultant supernatant was filtered (0.22-m, Millipore), then loaded onto a DEAE-Sepharose FF column (30 ml) (Amersham Biosciences) at 4°C. Elution employed a linear gradient of 150 -500 mM NaCl in 50 mM Tris-HCl, pH 8.0, and 1 mM EDTA. Fractions containing CarC were pooled, concentrated, and applied to an S75-Superdex column equilibrated in 150 mM Tris-HCl, pH 8.0, and 1 mM EDTA. Fractions containing CarC, judged to be ϳ90% pure (by SDS-PAGE), were pooled and concentrated. CarC was exchanged into 10 mM Tris-HCl, pH 8.0, using a PD-10 gel-filtration column (Amersham Biosciences) and concentrated to 40 mg/ml prior to crystallization (molecular masses measured by negative ion electrospray MS: 31,480 Ϯ 5 Da; calculated mass of CarC without N-terminal Met: 31,479.8 Da).
LC/MS-LC/MS was performed using a Waters high-performance liquid chromatography system connected to a Micro-Mass ZMD mass spectrometer in the negative ion mode. Assay mixtures as for the bioassays were mixed with MeOH (equal volume), chilled on ice for 10 min, and then centrifuged (17,800 ϫ g) for 5 min before analysis. Controls were carried out under identical conditions but with deactivated enzymes. The assay mixture (20 l) was injected onto a Synergi Polar-RP (250 mm ϫ 4.6 mm) column (Phenomenex) equilibrated in water at 1 ml/min. After 15 min a gradient to 90% MeOH was run over 5 min. These conditions were maintained for 5 min before returning to 100% water over 5 min and re-equilibration for 10 min.
Data Collection and Phase Determination-Diffraction data for Se-Met and L-NAP cocrystallized CarC were collected at 100 K on beamline 14.2 of the Synchrotron Radiation Source, Daresbury, UK with an ADSC Quantum 4 detector. The data were processed using MOSFLM and SCALA of the CCP4 suite (23) ( Table I). Fifteen selenium positions were located and refined with SOLVE (24). Phases were calculated from these positions with SHARP (25). Density modification and non-crystallographic symmetry averaging were carried out using DM of the CCP4 suite (23). 5% of the reflections were randomly selected to provide an R free test set.
Model Building and Refinement-Using the program O (26) all residues were built except for the N-terminal methionine, the C-terminal isoleucine, and two flexible loops linking ␣3/␤4 and ␤7/␤8 (the exact residues missing varied between monomers). Structure refinement was performed using REFMAC5 (27), with inclusion of the iron atoms, 2OG molecules, and 232 water molecules. There were no outliers in the Ramachandran plot (92.5%, 7.3%, 0.2% in the core, allowed, and generously allowed regions, respectively). Molecular replacement using AMoRe (28) was used to phase the data set collected with L-NAP cocrystals, and REFMAC5 was employed for the structure refinement. SCHEME 1. Carbapenem biosynthesis. Pathway a is consistent with the present work and that of Li et al. (39) and Bycroft et al. (14,21). For the possibilities of b see text. The CarC reaction is anticipated to require (minimally) 1 mol each of 2OG and O 2 and to produce 1 mol each of succinate, CO 2 , and H 2 O.
There was one Ramachandran outlier whose conformation appeared to be supported by the electron density (Leu-108 in subunit B) (92.8%, 7.0%, 0.0%, and 0.2% in the core, allowed, generously allowed, and disallowed regions, respectively). The electron density maps indicated that L-NAP was present in subunits B and C but at a much lower level, if at all, in A. Note that, although the submitted PDB file (1NX8) indicates one orientation (orientation I, see "Results and Discussion"), the orientation of L-NAP within the active site could not be unequivocally inferred from the electron density maps and a second orientation (II) is possible.
Molecular Modeling-Hydrogens were added to the complex using the HBUILD routine of CHARMM (version 27) (29). The final model included the protein atoms, the L-NAP ligand, 2OG, Fe(II), and 39 crystallographic waters, 4119 atoms in total. Models of the structure complexed with (3S,5S)-and (3S,5R)-carbapenams were prepared using QUANTA (30). Modeling studies were carried out with both carbapenams, using orientations I and II of L-NAP as initial "templates." In each case, a full occupancy oxygen atom was added (at a distance of 2.2 Å from Fe(II)) in the ligation position opposite to His-251 (the proposed catalytic cycle involves an Fe(IV)ϭO intermediate). Without this oxygen, the minimization resulted in the substrate carboxylate ligating to Fe(II). The involvement of such an iron-substrate carboxylate complex in catalysis seems unlikely both with respect to precedent (31) and on mechanistic grounds but cannot be ruled out. Each of these models contained 4119 atoms. A combined quantum mechanical/molecular mechanical (32, 33) (QM/MM) potential was used to perform minimizations of the model systems, while keeping Fe(II) and the atoms that are bound to it (His-101-N ⑀2 , Asp-103-O ␦1 , His-251-N ⑀2 , 2-keto, and 1-carboxylate oxygens of 2OG, and the additional oxygen atom in the two carbapenam systems) fixed in their original positions (Table II). In each case the ligand was described by a QM potential and the rest of the system by a coupled MM potential. The CHARMM standard all-atom parameters (34) were used in the MM region, except for the oxygen in the empty ligation position on Fe(II), which was given a charge of Ϫ1, and for the 2OG, whose charges were calculated by fitting them to the B3LYP/6 -31G* electrostatic potential in vacuum (35), using Gauss-ian98 (36). For non-bonded interactions, the electrostatics terms were truncated with a force switch function between 10 and 14 Å and the van der Waals terms with a shift function with a cutoff distance of 14 Å (37). The QM region was treated with the semi-empirical quantum mechanical method AM1 (38) implemented within CHARMM (33). The QM/MM minimizations included the steepest descent method followed by the Adopted Basis Newton-Raphson method implemented in CHARMM, until the average r.m.s. gradient was less than 0.01 kcal mol Ϫ1 Å Ϫ1 .  Energies and geometric parameters were obtained from the minimizations to compare the stability of the different models. The procedure was repeated for the (5R)-and (5S)-carbapenems (without the oxygen atom in the empty ligation position opposite to His-251) and for the two carbapenems except with 2OG replaced with succinate. In the latter case the results were almost identical to those obtained with 2OG.

RESULTS AND DISCUSSION
The CarC Reaction-The proposed carbapenam-3-carboxylic acid intermediate was prepared from the appropriate ␤-amino acid precursor using CarA. 2 The carC and carA genes were cloned from the E. carotovora genomic DNA and expressed in E. coli, and the corresponding proteins were purified by standard techniques. The (5S-carboxymethyl)-S-proline (trans-CMP) putative substrate for CarA was prepared via minor modification of reported methodology (39).
Assays with CarA alone, and combined CarA/CarC assays, containing the appropriate cofactors and potential CarA substrates, were conducted. With the trans-CMP substrate, no antibiotic activity was observed in assays with CarA alone. Assuming that CarA does not catalyze epimerization, the results imply that CarA can mediate ␤-lactam ring formation from trans-CMP to give (3S,5S)-carbapenam (Scheme 1, path a). They also suggest that the (3S,5S)-carbapenam can be converted by CarC to the (5R)-carbapenem. Because it cannot be entirely ruled out that the (3S,5S)-CMP used in this study was contaminated with a low level of its (3R,5R) enantiomer, on the basis of our data alone the possibility that the natural substrate for CarC is a (3R,5R) carbapenam cannot be discounted (Scheme 1, path b); however, this would be in conflict with the results and conclusions of both Li et al. (12) and Bycroft et al. (14,21). The level of substrate conversion effected by CarA with the trans-CMP was low compared with that of ␤-lactam synthetase (from the clavulanic acid biosynthesis pathway) with its natural substrate, possibly indicating an alternative in vivo substrate (see Scheme 1) or that a multiprotein complex is required to effect full activity (11). The organization of ␤-lactam biosynthesis proteins into a metabolon has also been suggested for clavulanic acid (40).
Crystallization and Oligomerization-Crystals of CarC complexed with Fe(II) and 2OG were obtained under anaerobic conditions. Crystals were also obtained anaerobically for CarC together with Fe(II), 2OG, and a substrate analogue (N-acetyl-L-proline) (see below). The structure was solved by molecular replacement using the model of SeMet-substituted CarC complexed with Fe(II).
Analysis of crystallographic symmetry revealed that CarC crystallizes as a hexamer comprised of two trimers (ABC and DEF in which A ϭ D, B ϭ E, and C ϭ F) (C222 1 ) (Fig. 1). Each asymmetric unit contains three monomers in a trimeric arrangement, with a hexamer being generated by a 2-fold crystallographic symmetry axis. Gel filtration and native gel electrophoresis studies also indicated that the predominant form of CarC in solution is also hexameric with low levels of monomeric and trimeric forms also being observed (molecular mass by analytical gel filtration: ϳ200 kDa).
Within each trimer the monomers are arranged such that the 2 M. C. Sleeman, unpublished results. active sites are well separated and directed toward the exterior of the hexamer, which possesses a large central channel. Hydrogen bonds and electrostatic interactions form links between the monomers and link the ABC and DEF trimers to form the hexamer; residues from ␣5 to ␣6 on the A subunit interact with the loop linking the ␤5 and ␤6 strands on the B-subunit.
With respect to interactions between the ABC and DEF trimers, A interacts most closely with E (the symmetry-equivalent of B), whereas B interacts with D (the symmetry-equivalent of A). C, however, interacts with F, its own symmetryequivalent. This difference results from a crystallographic 2-fold symmetry axis that runs between A/E and B/D pairs but between C and F monomers (Fig. 1b). In the case of the A/E interactions, residues from the N terminus, ␣2 and ␤8, from the A subunit interact with their counterparts on the E subunit to form the hexamer.
Overall Structure of the Monomer-The structures of the A, B, and C monomers are similar but not identical (the r.m.s. deviations of the C␣ atoms for the AB, BC, and CA pairs were 0.27, 0.26, and 0.17 Å, respectively). In the following discussions, the descriptions refer to the B monomer. The CarC main chain contains 14 ␤-strands, 8 of which (␤1-␤4, ␤6, and ␤11-␤14) combine to form the distorted double-stranded ␤-helix (DSBH or jellyroll) motif characteristic of the 2OG oxygenase superfamily that includes CAS (41), taurine dioxygenase (TauD) (42), and proline 3-hydroxylase (43) (Fig. 2). A DSBH is found in a wide range of metal binding (including the Cu(II)utilizing quercetin 2,3-dioxygenase (44)) and non-metal-binding proteins.
The presence of an extended insert (residues 135-225 (␣4 -␣6 and ␤7-␤10)) between the fourth and fifth strands of the DBSH places CarC within a distinct sub-group of 2OG oxygenases that includes CAS (45) but not deacetoxycephalosporin C synthase (DAOCS). The structural similarity between CarC and CAS (23% sequence identity) reinforces proposals that clavam and carbapenem biosynthesis have a close evolutionary relationship (8).

FIG. 3. Stereo view of the CarC active site showing 2OG binding site with bound L-NAP shown in yellow (monomer B). The Fe(II) is colored pink
and 2OG is pale blue. The 2mF o Ϫ DF c electron density "omit" map (red) is contoured at 1.5 , and the mF o Ϫ DF c electron density omit map (green) is contoured at 2 . TauD (42), although the presence of Arg-253 is distinct to CarC. The 2-keto-and 1-carboxylate groups of 2OG bind in a bi-dentate manner to the single iron, which is ligated by side chains from His-101, Asp-103, and His-251, all either part of or close to the DSBH (Fig. 3). These residues form a conserved 2-His-1-carboxylate triad of residues (46). CarC differs from CAS, but not TauD, in that its carboxylate ligand comes from an Asp rather than a Glu residue, highlighting the special nature of CAS in this regard. In the CarC⅐Fe⅐2OG⅐L-NAP structure, a water molecule is ligated to the ferrous iron opposite His-101 thus giving a six-coordinate arrangement. In the CarC⅐Fe⅐2OG structure, the 2OG is ligated such that its 2-ketogroup is opposite Asp-103, consistent with results for other 2OG oxygenases. The relative arrangement of Fe(II) ligands in the structures of anthocyanidin synthase and DAOCS positions the 2OG 1-carboxylate opposite His-251 (using the CarC numbering system); however, in TauD and CAS the 2OG 1-carboxylate is opposite His-101 (using the CarC numbering system). With CarC the 2OG 1-carboxylate appears to be in an intermediate position but is closer to being opposite to His-251 (angle for N⑀ His-251⅐Fe⅐1-carboxylate of 2OG, 150°; angle for N⑀ His-101⅐Fe⅐1-carboxylate of 2OG, 111°; for the L-NAP structure).
When a CAS⅐Fe⅐2OG⅐substrate complex was exposed to NO, acting as a dioxygen analogue, a rearrangement occurred placing the 2OG 1-carboxylate opposite His-251 (using the CarC numbering system), suggesting that in all cases dioxygen binding may occur trans to His-101 (31). However, as for related oxygenases, the substrates or substrate analogues (i.e. L-NAP for CarC) may not be correctly positioned to effect substrate oxidation if the ferryl species is formed opposite to His-101. It has been proposed that, following decarboxylation of 2OG, the ferryl species rearranges to be opposite His-251 and adjacent to the substrate (31). Alternatively, it cannot be ruled out that the 1-carboxylate can rearrange such that dioxygen can then bind opposite to His-251.
Substrate Binding-In addition to the Fe(II)⅐2OG complex and associated residues, the active site cavity comprises a largely hydrophobic region including Leu-98, Leu-106,  A 2OG turnover assay was used to screen a limited range of substrate analogues for binding to CarC. L-NAP and D-NAP were found to inhibit 2OG turnover (ϳ 60 and 25% of uncoupled turnover respectively). L-NAP was then used in cocrystallization experiments. The occupancy level of L-NAP differs between the three monomers A, B and C, from partial occupancy in B and C to very limited occupancy in A.
The CarC⅐Fe(II)⅐2OG⅐L-NAP structure is very similar to that of the CarC⅐Fe(II)⅐2OG complex. A water molecule is probably ligated to the ferrous iron opposite to His-101, possibly reflecting the observation that L-NAP inhibits (uncoupled) 2OG turnover. A strategy for potential inhibition of 2OG oxygenases is thus to hinder oxygen binding via stabilization of octahedral iron coordination chemistry. Due to the similar size of the acetyl and carboxylate groups of the L-NAP and its partial occupancy, it was not possible to unequivocally assign the orientation of L-NAP within the active site. Thus two possible orientations of L-NAP (I and II) related by a rotation of ϳ180°a re equally likely. In both orientations, two of the methylenes of L-NAP buttress against the side chain of Trp-202 in a hydrophobic interaction. In orientation I, the acetyl group is directed toward Arg-267 and Gln-269 and the backbone amide N-Hs of Gly-104. The hydrogens on the opposite face of the proline ring to the carboxylate are directed toward the iron. In orientation II, the carboxylate is directed toward the side chains of Arg-267 and Gln-269 and the backbone amide N-Hs of Gly-104. The hydrogens on the same face as the carboxylate are directed toward the iron. The proposed interaction of the carboxylate with Arg-267 in orientation II, is supported by the precedent of CAS, where the equivalent arginine (Arg-297, CAS) forms interactions with both the 1-carboxylate of 2OG and the substrate carboxylate, suggesting that this residue plays a key role in catalysis (Fig. 4).
Because the natural CarC substrate was not available in sufficient quantities for crystallization work, we used QM/MM modeling studies to investigate possible modes of carbapenam binding. Initial minimization of the model based on the CarC⅐Fe(II)⅐2OG⅐L-NAP crystal structure, in which the L-NAP is bound in orientation II resulted in a similar overall structure (r.m.s. deviation for non-hydrogen atoms, without waters: 1.5 Å), except for small changes in the loop between residues Gly-104 and Ser-109. Minimizations were then performed with the (3S,5S)-and the (3S,5R)-carbapenam models using both orientations I and II of L-NAP as initial "templates," i.e. by overlaying the pyrrolidine-carboxylate rings. Different binding possibilities are observed in the models, which suggest the Gly-104 to Ser-109 loop may be involved in binding a substrate carboxylate or lactam carbonyl. However, in each case, the results indicated similar energies for both the (3S,5S)-and the (3S,5R)-carbapenams (Fig. 5).
Mechanistic Discussion-Studies with other 2OG oxygenases have indicated that conformational changes occur upon substrate binding (47) such that the substrate binds to an "open" conformation but is isolated at the active site when it is oxidized. Thus the current crystallographic and modeled CarC structures may not precisely represent the active site conformation in which substrate binding and/or oxidation occurs. Nonetheless, they identify key residues involved in catalysis and suggest possible mechanisms for the CarC reaction.
Given the precedents with CAS and TauD, and the observation that the carboxylate of L-NAP may bind to the active site arginine (Arg-267), it is possible that the carboxylate of the (3S,5S)-carbapenam substrate binds to Arg-267 and Gln-269 during its oxidation, in a similar manner to that observed in orientation II of L-NAP and the associated (3S,5S)-carbapenam model; modeling suggests direct abstraction of a hydrogen at C-5 of the (3S,5S)-carbapenam may occur (Fig. 5). However in this case it would seem that re-orientation of the substrate would be required if a ferryl intermediate was to both abstract a hydrogen and re-hydrogenate "directly" at the C-5 position. The modeling studies indicate that a (3S,5R)-carbapenam, with or without a C-5 radical, could be accommodated in the active site, but there is no clear driving force for such a conformational change.
Instead, orientation I of L-NAP and associated models may represent the productive conformation. In this case epimerization may occur via abstraction of the C-3 or C-2 hydrogens (Scheme 2, a and b) or less likely, the C-1 hydrogen, followed by opening and ring closing of the bicyclic ␤-lactam system. Desaturation can then occur as in CAS. This proposal is attractive, because re-orientation of the substrate in the active site is not required for a ferryl intermediate to effect both epimerization and desaturation. In this proposal re-hydrogenation of a radical intermediate can lead to the (3S,5R)-carbapenam in a shunt pathway.
Other possibilities can be envisaged. Epimerization via abstraction at C-6 seems unlikely, because it would invoke a high energy primary radical and desaturation would either require significant re-orientation or unprecedented rearrangements/Hshifts. Given the precedent of ribonucleotide reductase (48) and others, the possibility that epimerization occurs via a process involving a protein based radical should be considered. However, analysis of the active site does not reveal clear candidates: SCHEME 2. Possibilities for the CarC-catalyzed epimerization/desaturation process. The proposed 5-endo trig (or 4-exo trig) radical cyclizations in a and b have synthetic precedent (50), including an imine substrate (51). Note the possible intermediacy of datively stabilized radicals. The proposed shunt pathway requires reduction of the ferryl intermediate (Fe(IV)ϭO 7 Fe(III)-O ⅐ ) to complete the catalytic cycle in a similar fashion to uncoupled cycles. Epimerization via hydrogen abstraction at C-1 in a process analogous to a is also a possibility, but the modeling studies suggest this is less likely.
Tyr-191 and Trp-202 were considered possibilities for such a role but appear incorrectly positioned. Thus, we favor a mechanism solely mediated via hydrogen transfers to the Fe(IV)ϭO and Fe(III)-OH intermediates, which are believed to occur with reactions catalyzed by related oxygenases involving rearrangements and desaturations, such as DAOCS (47) and CAS (41,49).